Fault Identification using Kernel Principle Component Analysis

In this paper, a new fault identification procedure based on Kernel Principal Component Analysis (KPCA) is proposed. In contrast to the linear PCA, a key disadvantage of the KPCA is that it cannot be used directly for fault identification, e.g. using contribution plots. Therefore, a fault identification strategy based on the power series approximation of the kernel functions is proposed. To demonstrate the functionality of the proposed algorithm, a nonlinear system is used. The proposed method is found effective in identifying the fault relevant variables for the individual and simultaneous occurrence of faults.